Research map / Adversarial robustness and security / Black-box and transfer attacks
Enhancing the Transferability of Adversarial Attacks Through Variance Tuning
Check your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| Add neighborhood gradient-variance tuning to MI-FGSM | MI-FGSM · ImageNet, Inc-v3 source → Inc-v4 | success rate (%): 43.6 -> 71.7 | Helped |
| Add neighborhood gradient-variance tuning to MI-FGSM | MI-FGSM · ImageNet, Inc-v3 source → Inc-v3_ens3 | success rate (%): 13.1 -> 32.8 | Helped |
| Add neighborhood gradient-variance tuning to NI-FGSM | NI-FGSM · ImageNet, Inc-v3 source → Inc-v4 | success rate (%): 51.7 -> 76.5 | Helped |
| Add neighborhood gradient-variance tuning to NI-CT-FGSM | NI-CT-FGSM · ImageNet, Inc-v3 source with CTM → Inc-v3_ens3 | success rate (%): 60.0 -> 78.6 | Helped |
| Add neighborhood gradient-variance tuning to NI-CT-FGSM | NI-CT-FGSM · ImageNet, nine defended models; ensemble source | average success rate (%): 83.5 -> 90.1 | Helped |
| Add neighborhood gradient-variance tuning to NI-CT-FGSM | NI-CT-FGSM · ImageNet, nine defended models; Inc-v3 source | average success rate (%): 52.8 -> 70.0 | Helped |
| Set neighborhood size factor β to 0, removing variance tuning (ablation) | VMI-FGSM and VNI-FGSM · ImageNet black-box models | transferability (success rate) | Hurt |
| Set neighborhood sample count N to 0, removing variance tuning (ablation) | VMI-FGSM and VNI-FGSM · ImageNet black-box models | transferability (success rate) | Hurt |
| Increase neighborhood size factor β (ablation) | VMI-FGSM and VNI-FGSM · ImageNet black-box models | transferability (success rate) | Mixed |
| Increase sampled neighborhood examples beyond N=20 (ablation) | VMI-FGSM and VNI-FGSM · ImageNet black-box models | transferability (success rate) | No clear effect |
Closest papers by meaning
- Enhancing Adversarial Transferability with Adversarial Weight Tuning (AAAI 2025 · similarity 0.91)
- Improving Transferable Targeted Attacks with Feature Tuning Mixup (CVPR 2025 · similarity 0.89)
- Improving Transferable Targeted Adversarial Attacks with Model Self-Enhancement (CVPR 2024 · similarity 0.88)
- Ensemble Diversity Facilitates Adversarial Transferability (CVPR 2024 · similarity 0.88)
- Boosting the Transferability of Adversarial Attack on Vision Transformer with Adaptive Token Tuning (NeurIPS 2024 · similarity 0.88)
- Towards Transferable Adversarial Attacks with Centralized Perturbation (AAAI 2024 · similarity 0.87)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.